DictateAnywhere

Hybrid voice dictation for Windows — types in any active text field using local offline faster-whisper or optional Azure/Sarvam AI cloud backends.

📥 Download Releases • 📖 Documentation

---

Features

Feature	Details
Hybrid STT	Offline faster-whisper (CPU/GPU) + Azure Speech / Sarvam AI cloud fallback
GPU Accelerated	High-performance transcription via CUDA/cuBLAS for NVIDIA RTX GPUs
Real-Time Previews	Live text appears in the overlay as you speak (sub-second latency)
Sarvam AI Support	High-performance STT for Indian languages via WebSocket streaming
LLM Text Polishing	Integrated with local Ollama to automatically fix grammar, rewrite, or act as a conversational assistant
Always available	System tray icon — lives quietly in your taskbar
Floating mic button	Draggable, always-on-top toggle with modern, crisp icons for all states
Countdown ring	Sweep arc on the mic button shows remaining recording time (green → amber → red)
Global hotkey	Configurable (default Ctrl+Alt+D), toggle or push-to-talk
Types anywhere	Injects text at your cursor in any Windows app
Transcription preview	Floating dark overlay shows live tentative text and last 3 dictated lines
Live level meter	smooth live waveform visualizer in the preview overlay shows mic amplitude
Spoken punctuation	"period" → . "comma" → , "new line" → ↵ etc.
Auto-capitalisation	Capitalises after sentence endings automatically
Word corrections	Persistent find-and-replace rules applied after every transcription
Session history	Searchable log of every dictated utterance — copy, export, or clear
Auto-update checker	Silently checks GitHub Releases at startup; notifies when a new version is available
VAD filtering	WebRTC Voice Activity Detection — CPU only active while you speak
Secure key storage	API keys stored in Windows Credential Manager (DPAPI)
Model selector	tiny / base / small (recommended) / medium / large
Multi-language	20+ languages via Whisper; specialized Indian support via Sarvam
Start with Windows	Optional registry key entry
App Launcher	Bind custom voice triggers (e.g. "open notepad") to run local applications
Profile Sync	Export and Import settings configurations to/from a JSON file
Premium Branding	Glowing neon mic window and tray icons with live status indicator dots
Standalone .exe	Build with PyInstaller — no Python install needed for end-users

---

Quick Start

Requirements

• Windows 10 / 11 (x64)
• Python 3.11, 3.12, or 3.13 — python.org/downloads
• A microphone

Install

git clone https://github.com/RhythmicDias/DictateAnywhere.git
cd DictateAnywhere

scripts\create_venv.bat
scripts\install.bat

Run

scripts\run.bat

DictateAnywhere starts silently and appears in the system tray (bottom-right of your taskbar).

---

Usage

Action	How
Start / stop dictation	Press Ctrl+Alt+D (configurable)
Start / stop dictation	Click the floating mic button
Start / stop dictation	Right-click tray icon → Start / Stop
Open settings	Right-click tray icon → Settings…
View session history	Right-click tray icon → Session History
Toggle preview overlay	Right-click tray icon → Toggle Preview
Move floating button	Click and drag it anywhere on screen
Quit	Right-click tray icon → Quit

Transcription preview overlay

While recording, a dark floating bar appears at the bottom of your screen showing:

• ● Listening… status with a smooth live waveform visualizer to show your microphone activity
• The last three dictated lines (newest is white; older lines dim)

The overlay auto-hides a few seconds after dictation ends. You can drag it anywhere and close it with ✕. The hide delay is configurable in Settings → Advanced.

Countdown ring on the floating button

When recording is active, a sweep arc appears inside the mic button tracking how much of the maximum recording time has been used:

Ring colour	Meaning
Green	> 50% of time remaining
Amber	20–50% of time remaining
Red	< 20% remaining — wrapping up soon

A small seconds counter appears at the top of the button. The ring and counter disappear instantly when recording ends.

Spoken punctuation commands

Say	Gets typed
"period" / "full stop"	.
"comma"	,
"question mark"	?
"exclamation mark"	!
"new line"	line break
"new paragraph"	blank line
"semicolon"	;
"colon"	:
"open quote" / "close quote"	"
"dash"	—
"ellipsis"	…
"delete that" / "scratch that"	removes the spoken word

Word corrections

Open Settings → Corrections to define find-and-replace rules applied after every transcription.

Column	Description
Find	The word or phrase Whisper tends to get wrong
Replace	What you actually want typed

Rules are case-insensitive by default. They are stored in %APPDATA%\DictateAnywhere\corrections.json and apply on top of Whisper's output after spoken punctuation processing.

Example rules:

Find	Replace
colour	color
Starbucks	Starburst
gonna	going to

---

Configuration

Open Settings from the tray icon menu. Changes take effect immediately.

Engine tab

• Engine mode — hybrid (recommended), local, or cloud
• Whisper model size — tiny is the fastest; small is the best CPU/accuracy tradeoff
• Compute type — int8 (efficient) or float16 (fastest for GPU)
• Local device — cuda (recommended for NVIDIA) or cpu
• Language — BCP-47 code (en, fr, de, auto, …)

Audio tab

• Choose microphone device
• VAD aggressiveness (0–3)
• Silence timeout before auto-stop
• Max recording length

Hotkey tab

• Set any key combination (e.g. ctrl+alt+d, f9, ctrl+shift+space)
• Choose between toggle and push-to-talk modes

Floating Button tab

• Show / hide the floating button
• Size and opacity
• Always-on-top toggle

Preview tab

• Enable or disable the transcription preview overlay
• Configure the auto-hide delay (ms)
• Opacity & Color — customize the overlay's transparency and text color for better visibility

Corrections tab

• Add, edit, and remove word correction rules
• Rules are applied after every transcription in the order listed

App Launcher tab

• Map spoken commands (e.g. "open notepad") to specific local application executable files to launch them instantly.

Azure Cloud tab

• Azure Speech API key (stored securely in Windows Credential Manager)
• Azure region
• Test connection button

Advanced tab

• Toggle spoken punctuation
• Toggle auto-capitalisation
• Text injection method (clipboard or SendInput)
• Start with Windows
• Backup & Sync — "Export Settings" and "Import Settings" buttons at the bottom of the dialog box to save or load JSON profiles.
• Log level
• Updates section — enable/disable automatic update checks; "Check now" button for an on-demand check

---

AI Text Polishing (Ollama & Gemini)

Transform your speech in real-time using Large Language Models.

Local via Ollama

1. Install Ollama
2. Download a model (e.g., ollama run llama3)
3. In DictateAnywhere Settings → Polish, select ollama and pick your model.

Cloud via Google Gemini

1. Get an API key from Google AI Studio
2. In DictateAnywhere Settings → Cloud STT, paste your Gemini API key.
3. In Settings → Polish, select gemini and choose a model (e.g., gemini-flash-lite-latest).

Translation & Custom Prompts

Use the Custom Prompt action to turn DictateAnywhere into a real-time translator or specialized assistant.

• Example Prompt: "Translate the following text to Spanish and output ONLY the translation."
• Example Prompt: "Format the following text as a SQL query."

---

Google Gemini (Cloud STT)

DictateAnywhere supports Google's Gemini models for highly accurate, multi-language cloud transcription.

1. Get a free-tier API key from Google AI Studio.
2. Paste the key in the Cloud STT tab.
3. Switch your Engine Mode to gemini in the Engine tab.

---

Azure Speech (optional cloud backend)

The free tier of Azure Speech gives you 5 hours of transcription per month at no cost.

1. Create a free Azure account at azure.microsoft.com/free
2. Create a Speech resource (Free tier F0)
3. Copy your Key 1 and Region
4. In DictateAnywhere Settings → Azure Cloud tab, paste the key and set the region
5. The key is encrypted by Windows and never touches disk in plain text

---

Sarvam AI (Indian Language specialized STT)

For the best experience with Hindi, Malayalam, Tamil, etc., DictateAnywhere integrates with Sarvam AI.

1. Get an API key from the Sarvam AI Dashboard.
2. In DictateAnywhere Settings → Sarvam AI tab:
   • Paste your API Key (stored securely in Windows Credential Manager).
   • Select your preferred model (e.g., saaras:v3).
   • Enable WebSocket Streaming for ultra-low latency real-time transcription.
3. Switch the Engine Mode to sarvam in the Engine tab.

---

Auto-update checker

DictateAnywhere silently checks GitHub Releases at startup (after a 15-second delay) and shows a notification if a newer version is available.

• At most once per day — subsequent launches that day skip the check
• Three choices when a new version is found:
  • Download — opens your browser to the GitHub release page
  • Skip this version — suppresses notifications for that specific release (persisted to config)
  • Remind me later — dismisses; will show again on the next daily check
• Manual check — Settings → Advanced → Updates → Check now
• Disable entirely — uncheck "Check for updates automatically"

---

Session history

Right-click the tray icon → Session History to open a log of every utterance dictated in the current run.

Control	Action
Search box	Filter utterances by text
Copy	Copies selected entry to clipboard
Export	Saves full history as a .txt file
Clear	Wipes the history list

History is in-memory only and resets on restart. Use Export to save anything you need to keep.

---

Build a standalone .exe

No Python required on the target machine after building.

scripts\build.bat

Output: dist\DictateAnywhere\DictateAnywhere.exe

---

Development

# Activate the venv
.venv\Scripts\activate

# Run tests
scripts\test.bat

# Run in dev mode (console visible)
scripts\run_dev.bat

# Lint
pip install ruff
ruff check src\

Project layout

DictateAnywhere/
├── src/dictateanywhere/
│   ├── main.py                  ← entry point & orchestrator
│   ├── __init__.py              ← version string
│   ├── audio/
│   │   ├── capture.py           ← mic input, RMS level callback, TimedCapture
│   │   └── vad.py               ← WebRTC voice activity detection
│   ├── transcription/
│   │   ├── engine.py            ← abstract STT base class
│   │   ├── local_engine.py      ← faster-whisper (offline)
│   │   └── cloud_engine.py      ← Azure Speech SDK (cloud)
│   ├── core/
│   │   ├── hotkey_manager.py    ← global hotkey registration
│   │   ├── text_injector.py     ← types text at cursor (clipboard / SendInput)
│   │   ├── punctuation.py       ← spoken → symbol conversion
│   │   ├── corrections.py       ← word correction rules (corrections.json)
│   │   └── updater.py           ← GitHub Releases update checker
│   ├── ui/
│   │   ├── tray.py              ← system tray icon (pystray)
│   │   ├── floating_widget.py   ← draggable mic button + countdown ring
│   │   ├── preview_window.py    ← transcription overlay + level meter
│   │   ├── history_window.py    ← session history viewer
│   │   └── settings_window.py   ← tabbed settings dialog (tkinter)
│   └── utils/
│       ├── config.py            ← JSON config in %APPDATA%\DictateAnywhere\
│       └── secure_storage.py    ← API key → Windows Credential Manager
├── tests/                       ← pytest test suite
├── scripts/                     ← install, run, build, test .bat files
├── assets/icons/                ← .ico / .png for tray and .exe
├── .github/workflows/           ← CI + release workflow
├── requirements.txt
├── pyproject.toml
└── README.md

Key data files (per-user, %APPDATA%\DictateAnywhere\)

File	Contents
config.json	All settings (hotkey, engine, UI prefs, update state)
corrections.json	Word correction rules
dictateanywhere.log	Rolling application log

---

Roadmap

• Per-app language profiles (e.g. English for Word, French for LibreOffice)
• Multi-monitor floating widget awareness
• macOS / Linux support (pynput instead of pywin32)
• Real-time streaming transcription (Local & Sarvam WebSocket)
• GPU acceleration support (CUDA via CTranslate2)
• Noise floor auto-calibration
• Custom wake word to start recording hands-free

---

Contributing

Pull requests are welcome! Please:

1. Fork the repo and create a feature branch
2. Run scripts\test.bat and make sure all tests pass
3. Follow the existing code style (type hints, docstrings, no print(), use the logger)
4. Open a pull request with a clear description

---

License

MIT — see LICENSE.

---

Acknowledgements

• faster-whisper — Optimised Whisper inference by SYSTRAN
• OpenAI Whisper — The original speech recognition model
• Azure Cognitive Services Speech
• pystray — System tray icon library
• webrtcvad — WebRTC voice activity detection
• sounddevice — PortAudio bindings for Python